New Zealand recorded its lowest-ever average somatic cell count in 2023/24. That is a herd average. From 1 January 2027 every dry-cow antibiotic has to be justified for the individual animal. The individual is where the average tells you nothing.
Once a year, at the end of lactation, every cow in the herd gets sorted into one of three piles. Antibiotic dry-cow therapy. Internal teat sealant only. Or culled. From the 2027 season the vet has to be able to justify that first pile animal by animal, in writing. The sorting still happens in an afternoon, in a yard, against a number that was measured weeks ago.
It is a decision that cannot be revisited. The cow is dry for weeks; whatever was in the udder at dry-off has the whole dry period to resolve or to establish. You find out how the sort went at calving, when she either milks clean or she does not.
Get it wrong either way and you pay: in antibiotics you did not need, or in a lactation you cannot get back.
both errors, and they run in opposite directionsOne cow, at the last herd test of the season, reads 170,000 cells per millilitre. Under the rule she is comfortably below the cut-point, so she gets a teat seal and no antibiotic, and the sort moves on to the next cow.
Underneath that one number are four quarters. Three of them are clean at around 60,000. The fourth is running at 500,000 and has an established infection in it. A herd-test sample is a composite: milk from all four quarters, mixed, measured once. Average the four and you get 170,000, and the infected quarter disappears into the three that are not.
The sampling makes it worse. Herd tests are roughly monthly, and irregularly spaced. What you hold is a handful of points on a line that moves continuously: a level, taken on a day, standing in for a process. Between her last test and dry-off she may have cleared it, held it, or seeded two more cows at every milking since.
And the thing you actually need is not on the list at all. The question at dry-off is not is she infected. It is will the antibiotic work on her. A chronic infection in an older cow, well established, may not clear no matter what you put in it. The dose is spent and the cow is no better. Nothing in a somatic cell count distinguishes the cow who will cure from the cow who will not. It was never measuring that.
She is sealed. She calves. In the second week she is clinical, and the record will show that she was under the threshold at her last test, which is true, and which explains nothing.
The Veterinary Council of New Zealand’s updated prescribing advice for dry cow therapy comes into effect on 1 January 2027.
Under it, veterinarians must be able to justify each individual animal’s treatment by ensuring there is sufficient evidence to support a need for DCAT use for each animal for which the product is authorised
, meaning individual cow somatic cell count results, and must record the decision-making process, the supporting evidence, and their reasons for using the antibiotic in each cow. Whole-herd treatment survives only where the herd meets both bulk-milk criteria: an arithmetic mean bulk milk SCC of at least 200,000 cells/mL through the seventh month of lactation, and a monthly increase of more than 40,000 cells/mL between the sixth and seventh. Practices are already writing to clients about what that means: one Waikato clinic told its farmers in March 2026 that from 1st January 2027 there will be penalties for vets who prescribe DCT without justification.
DairyNZ puts the share of New Zealand herds that meet the bulk-milk threshold justifying blanket treatment at fewer than 5%. At Taranaki Veterinary Centre, a practice a 2023 Ministry for Primary Industries audit singled out as demonstrating the highest standard of stewardship of prudent antibiotic use
, 19% of herds still receive whole-herd DCAT. Getting there took that clinic from treating 80% of adult cows in 2014 to 50% now, over more than a decade, with two years of staff training in how to have the conversation. Those are different populations and not a like-for-like ratio. They are still the shape of the problem: the best in the country, still four times the qualifying rate, seventeen months out.
What that changes is the nature of the constraint. The scarce thing at dry-off is not the drug. You can buy as much of that as you like. The scarce thing is permission to use it, and from 2027 permission is granted one animal at a time, against evidence, on the record. That is an unusual budget: it does not run out, it has to be earned, and the earning has to be written down.
And the dose is not the end of it. A cow given antibiotic dry-cow therapy carries a withholding obligation that runs past calving: the New Zealand Veterinary Association’s position on bobby calves requires strict application of the milk withholding periods for DCT rather than a general clean-milk rule, and calves treated directly with antimicrobials are disqualified from bobby calf slaughter altogether. So an unnecessary dose does more than spend permission and seed resistance. It attaches a tracking obligation to one animal, months later, in a trade that is watched from outside. A cow you treat in May is still a constraint in August.
It is also, on the Taranaki evidence, not a switch. That clinic moved by a few percentage points a year, farm by farm, against reasons for resisting that were mostly not financial. A farmer who once lost cows to a teat-seal failure will not be moved by an argument about cost. What changes a mind at that scale is being able to say, for this cow, on this evidence, here is what we expect to happen.
Which makes the honest framing of this the opposite of the usual one. New Zealand is not in a cell-count crisis. The national average is the best it has ever been. The problem is that a national average has never once told a vet which cow in front of them should get the needle.
An average that is improving and a decision that is getting harder are not a contradiction. They are the same fact seen from two ends.
The rule is a cell-count threshold. Treat cows above a cut-point at the last herd test, or with a clinical history; seal the rest. DairyNZ’s Technote 14, dated January 2025, sets it out plainly: In the past, thresholds of 150,000 cells/mL have been used. As herds implement better mastitis management in lactation, the threshold can be increased to 200,000 or 250,000 cells/mL.
The rule is strong, and worth setting out that way. It is free, it runs on data the farm already collects and already pays for, and it produces exactly the audit trail the new authorisation requirement asks for. Every vet and every farmer in the country understands it. And it works: culture- or algorithm-guided selective dry-cow therapy cut quarter-level antibiotic use at dry-off from 100% under blanket treatment to 45%, across seven herds with an internal teat sealant in every quarter, and the companion trial found either programme unlikely to increase clinical mastitis risk
in the following lactation (Rowe et al., two papers, Journal of Dairy Science, 2020). Roughly half the antibiotics, with no measurable udder-health penalty. That is the bar.
And the same document that recommends the threshold reports how well it separates. When using a maximum cow SCC of >200,000 cells/mL as the cut-point, for the last herd test within 80 days before drying off, the Se and Sp were 0.78 and 0.79, respectively.
About one infected cow in five is below the line and gets sealed. About one uninfected cow in five is above it and gets an antibiotic she did not need. Those are not our numbers, and they are not an attack on the rule. They are the rule’s own published operating characteristics, in the guidance that recommends it. DairySmart’s field data points the same way from the other side: approximately 20% of high-SCC cows grow no bugs.
The threshold does not sort badly. It sorts on the wrong quantity, about as well as that quantity allows.
the gap, stated at its narrowestSo the bar is not beat blanket treatment. Selective dry-off already did that, and from 2027 the authorisation requirement makes it the default. The bar is this. Does a forecast beat a good threshold rule on the two errors the threshold makes and publishes? And does it do so while producing the per-cow justification the authorisation requires anyway?
Forty cows, ordered left to right by their last herd test, with the cut-point drawn across. The rule sorts on height. But infection is not height, and the two only agree about four times in five. So infected cows sit below the line and uninfected cows sit above it, in the proportions the guidance itself reports. Underneath, the reason one of those errors is structural: a herd test is a composite of four quarters, and one badly infected quarter averaged with three clean ones lands under the cut-point without ever appearing.
Drawn at the sensitivity and specificity DairyNZ publishes for a >200,000 cells/mL cut-point (0.78 and 0.79). Herd size and the share of cows infected are illustrative; the error proportions and the quarter arithmetic are not.
Among the cows the threshold already selects, which will actually clear if treated and which will carry the infection through the dry period regardless. This is the quantity the decision turns on, and the one no cell count measures. It also decides whether a dose is spent well, which is what the authorisation asks the vet to assert.
Cow-level inference from sparse, irregularly spaced, whole-udder samples, pooled across cows within a herd and across herds, so that a single cow with four tests borrows strength from thousands with the same shape. The target is the cows the rule misses: below the cut-point, infected, sealed anyway.
The constraint here is not money or supply. It is that each dose must be justified for the individual animal, on evidence, in the record. An optimiser that produces the ranked action list and the reasoning behind each call is producing the compliance artefact as a by-product of the decision, rather than as a second job afterwards.
Lift over a well-set cut-point, reported with uncertainty, on both axes at once: doses used, and udder health in the following lactation. Walked forward through seasons already lived, using only the data that existed on each dry-off date. If the threshold captures most of the value, that is the answer and it should surface in the back-test rather than in year two.
That New Zealand has a somatic cell count crisis. It does not. The national average is the lowest it has ever been, and saying otherwise to this industry would be both wrong and obvious. The argument is about the variance the average hides, and about a decision the average was never able to make.
That the Rowe trials are a New Zealand result. They were run in the United States. They establish the solution class and set the bar; nothing here claims a New Zealand result.
That the 19% and the 5% are the same measurement. One is the share of herds at a single named clinic receiving whole-herd DCAT; the other is DairyNZ’s estimate of the share of herds nationally meeting the bulk-milk criteria. They are quoted as they were published and are not divided into one another.
A pilot runs your own dry-off decisions again across seasons you have already lived, scored against what actually happened in the lactation that followed. You already know how those seasons went. That is what makes it a test rather than a demonstration.